Based on the generalized cross-correlation method (GCC), the period peaks in the cross-correlation function (CCF) of frequency shift keying (FSK) signals will degrade the estimation performance of time difference of arrival (TDOA), as well as the estimation performance of frequency difference of arrival (FDOA), through the cross ambiguity function (CAF), in the case of a low signal-to-noise ratio (SNR). An improved TDOA/FDOA estimation algorithm made using period peaks is proposed in this paper to better understand the performance of TDOA/FDOA estimation for FSK signals. First, the cross ambiguity function of FSK signals is computed, and the peak position of the CAF is found to obtain coarse TDOA/FDOA estimation results, as per the usual method. Next, the mainlobe and period sidelobes are found according to the peak position and period interval; then, the peaks of each sidelobe around the mainlobe are found, and the period TDOA estimations are obtained. Then, the improved TDOA estimation can be used to calculate the average value of period TDOA estimations. Last, the period sidelobes accumulate to the mainlobe, and the improved FDOA estimation result are obtained by finding the peak position of the accumulated mainlobe. Simulations are performed to demonstrate that the proposed algorithm provides a better performance.